CN115659833A - Power network node vulnerability assessment method based on BP neural network - Google Patents
Power network node vulnerability assessment method based on BP neural network Download PDFInfo
- Publication number
- CN115659833A CN115659833A CN202211396612.8A CN202211396612A CN115659833A CN 115659833 A CN115659833 A CN 115659833A CN 202211396612 A CN202211396612 A CN 202211396612A CN 115659833 A CN115659833 A CN 115659833A
- Authority
- CN
- China
- Prior art keywords
- network
- node
- power
- matrix
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 53
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000012360 testing method Methods 0.000 claims abstract description 120
- 238000012549 training Methods 0.000 claims abstract description 106
- 239000011159 matrix material Substances 0.000 claims abstract description 102
- 239000013598 vector Substances 0.000 claims abstract description 34
- 238000010606 normalization Methods 0.000 claims abstract description 22
- 238000013075 data extraction Methods 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 12
- 230000005540 biological transmission Effects 0.000 claims description 52
- 230000009467 reduction Effects 0.000 claims description 46
- 238000003062 neural network model Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 22
- 238000011176 pooling Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 12
- 239000011541 reaction mixture Substances 0.000 claims description 12
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 6
- 238000012546 transfer Methods 0.000 claims description 6
- 238000010801 machine learning Methods 0.000 claims description 5
- 210000002569 neuron Anatomy 0.000 claims description 5
- 230000015556 catabolic process Effects 0.000 claims description 4
- 238000006731 degradation reaction Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 3
- 238000002759 z-score normalization Methods 0.000 claims description 2
- 239000000284 extract Substances 0.000 abstract 1
- 238000011160 research Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- FESBVLZDDCQLFY-UHFFFAOYSA-N sete Chemical compound [Te]=[Se] FESBVLZDDCQLFY-UHFFFAOYSA-N 0.000 description 1
- 238000011425 standardization method Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a power network node vulnerability assessment method based on a BP neural network, which comprises the following steps: building an electric power training network and an electric power testing network model; acquiring an electrical characteristic matrix; constructing a characteristic data extraction algorithm; acquiring a vector dimension unified electrical characteristic matrix; acquiring an electrical abstract feature matrix of a power network node; defining a node electrical characteristic label; constructing a node characteristic attribute; constructing a data set to be normalized and normalizing; building a BP neural network training model; inputting the power test network into a BP neural network training model to obtain a test value after inverse normalization processing; acquiring a priority protection node set and applying key protection; the method comprehensively considers the topological structure and the electrical characteristics of the power network, extracts the electrical characteristic dimensions to be uniform, is suitable for high-dimensional and nonlinear large-scale complex power grid vulnerability assessment, has universality for all power grids, quickly and accurately obtains the priority protection set, and can effectively improve the safe and stable operation efficiency of the power system.
Description
Technical Field
The invention belongs to the technical field of safe and stable operation of electric power systems, and particularly relates to a power network node vulnerability assessment method based on a BP neural network.
Background
With the rapid development of the global energy internet, the power network plays a crucial role in its development as the most important component of the energy internet. Therefore, weak links and potential dangers in the operation of the power system are analyzed, the vulnerability assessment of the power grid is obtained, the prevention and protection of the vulnerable units are enhanced, and therefore the safety and stable operation of the power system are improved, and the method has important significance. With the continuous expansion of the scale of the power grid, the power grid becomes a large-scale complex artificial network with high dimension and nonlinearity, and the data volume in the database of the power system is greatly enriched, so that the power system enters a big data era. How to effectively utilize the mass data and mine valuable information from the mass data to serve for vulnerability assessment of the power network becomes a brand-new field of vulnerability research.
At present, the research on the vulnerability mechanism of the power system mainly comprises structural vulnerability and state vulnerability, and a structural vulnerability assessment index and a state vulnerability assessment index are provided according to different research mechanisms. The research on the vulnerability of the structure of the power system is mainly based on a complex network theory and researches the inherent vulnerability of the topological structure of the power grid; the state vulnerability assessment of the power system mainly comprises transient stability analysis and an assessment method based on probability theory, and the change condition of each state parameter when the power grid operates is mainly considered.
In the current method for evaluating the vulnerability of the power system, the extraction of the basic characteristic data of the power system for determining the vulnerability usually depends on manual experience, and the uncertainty of the effect of manually extracting the node characteristics exists; the vulnerability assessment of the power network nodes depends on the established mathematical model, the adopted basic characteristics of the power system and the accuracy of model parameters. However, the actual power system is a complex nonlinear system, and various physical limitations and saturation constraints make it difficult to build an accurate mathematical model. With the expansion of the operation scale and the increase of complexity of the power grid, the traditional power grid vulnerability assessment model lacks accuracy and scientific guidance, and cannot meet the requirement of quickly and accurately assessing the vulnerability of the power network nodes.
Therefore, the method has important theoretical and academic significance and engineering application significance for promoting the practicability of the electric network node vulnerability assessment method and is also an important practice for applying the next generation information technology to the safety protection of the electric power system.
Disclosure of Invention
In order to overcome the technical defects, the invention provides a power network node vulnerability assessment method based on a BP neural network.
The purpose of the invention can be realized by the following technical scheme:
the method for evaluating the vulnerability of the power network node based on the BP neural network specifically comprises the following steps:
s1: respectively building an electric power training network topological structure model comprising a plurality of independent electric power networks and an electric power testing network topological structure model comprising a single independent electric power network;
s2: extracting network topology characteristics of each independent power network in the power training network and network topology characteristics of the power testing network based on a complex network theory, wherein the network topology characteristics comprise node values, node betweenness centrality, node approximate centrality, node feature vector centrality, node clustering coefficients and node types;
s3: acquiring an electrical characteristic matrix in each independent power network and a power test network in a power training network, wherein the electrical characteristic matrix comprises a power transmission distribution factor matrix, an electrical distance matrix and a maximum transmission capacity matrix;
s4: calculating the network efficiency of each independent power network in the power training network, and calculating the network efficiency reduction rate after any node in the independent power networks is removed;
s5: adopting a convolution kernel and a pooling window in an asymmetric form in machine learning, and constructing a feature data extraction algorithm according to the dimension of the electrical feature matrix of each independent power network and each power test network in the power training network;
s6: acquiring a row vector dimension unification electrical characteristic matrix of each independent power network and each independent power test network in the power training network according to the characteristic data extraction algorithm;
s7: normalizing the vector dimension unified electrical characteristic matrix to obtain an electrical network node electrical abstract characteristic matrix corresponding to each independent power network and the power test network with unified row vector dimension and data normalization, wherein the electrical abstract characteristic matrix of the power network node comprises a power transmission distribution factor abstract characteristic matrix, an electrical distance abstract characteristic matrix and a maximum transmission capacity abstract characteristic matrix;
s8: extracting elements in row vectors of the electrical abstract feature matrix of the power network nodes, and defining the elements as node electrical feature labels in each independent power network and power test network in the power training network;
s9: according to network topological characteristics and node electrical characteristic labels of each independent power network and each power test network in the power training network, node characteristic attributes of each independent power network and each power test network in the power training network are constructed;
s10: removing any node in the independent power network in the S4 to obtain the network efficiency reduction rate, wherein the electrical abstract feature matrix of the power network node in the S7 comprises a power transmission distribution factor abstract feature matrix, an electrical distance abstract feature matrix and a maximum transmission capacity abstract feature matrix, and the network topology characteristics in the node feature attribute in the S9 comprise a node value, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficients and a node type to form a data set to be normalized;
s11: normalizing each item of data in the data set to be normalized in S10;
s12: the method comprises the following steps of building a BP neural network training model, wherein the BP neural network comprises an input layer, a hidden layer and an output layer, and the concrete steps of building the BP neural network training model comprise:
s121: determining the number of neurons of each layer of an input layer, a hidden layer and an output layer in the BP neural network;
s122: taking the node electrical feature labels and the normalized node feature attributes in the S11, including node values, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficients and node types as input parameters of a BP neural network input layer;
s123: taking the normalized network efficiency reduction rate of the independent power network where the node is located after the node is removed in the S11 as an output parameter of the BP neural network output layer;
s124: inputting the training set into a BP neural network, establishing nonlinear mapping between the characteristic attribute of the input layer power network node and the network efficiency reduction rate after the output layer node is removed, and completing the establishment of a BP neural network training model;
s13: inputting the power test network into the BP neural network training model established in S12, obtaining a test result of the network efficiency reduction rate after the power test network node is removed, and performing reverse normalization on the test result to obtain a test value of the network efficiency reduction rate after the node is removed after the reverse normalization processing;
s14: and establishing a power test network node vulnerability set according to the test value of the network efficiency reduction rate of the power test network, performing descending sorting according to the network efficiency reduction rate of all elements in the node vulnerability set, adding the nodes with the top rank to a key protection node set to obtain a priority protection node set, and applying key protection to the nodes in the set.
Furthermore, generators, loads and substations in the electric power training network topology model and the electric power testing network topology model are abstracted as nodes in the network topology model, and power transmission lines and transformer branches in the electric power training network topology model and the electric power testing network topology model are abstracted as edges in the network topology model, which can be expressed as follows:
in the formula (1), the reaction mixture is,
n independent power networks are selected as sample data in the power training network, the power training network topological structure model comprises N independent power network topological structure models,
set representing all node types in the kth independent power grid in the power training network topology model, including generator node set
Load node set
Transformer substation node set
Representing all edge sets in the power training network topological structure model;
the power test network topology model can be represented as:
G P ＝(V P ,E P ) (2)
in the formula (1), G p Showing the selection of a single independent power network as sample data in a power test network, V P Representing a set of all node types in a power test network topology model, including a set of generator nodes
Load node set
Transformer substation node set
E P Representing all edge sets in the power test network topology model.
Further, the network efficiency of each independent power network in the power training network is calculated as the network efficiency of the kth independent power grid
For example, the calculation process is shown in formula (3):
in the formula (3), the reaction mixture is,
for the total number of kth individual grid generator nodes in the power training network,
for the kth individual grid load node total in the power training network,
for generator node k _ g and load node k _ _inthe kth independent power grid in the power training networkl is the electrical distance between the electrodes,
the transmission capacity between the generator node k _ g and the load node k _ l for the power training network;
calculating the network efficiency reduction rate after removing any node in the independent power network and the network efficiency reduction rate after removing the node i from the kth independent power network
For example, the calculation process is shown in formula (4):
in the formula (4), the reaction mixture is,
to remove the network efficiency of the independent power network in which the node k i is located,
the larger the node k _ i is, the more influence on the power network where the node k _ i is located is, and the more important the node is.
Further, the calculation process of the electrical distance between the generator node k _ g and the load node k _ l in the kth independent grid in the electric power training network is as shown in formula (5):
in the formula (5), Z k_gg Is the k-th independent power grid node impedance matrix k _ g row and k _ g column elements, Z in the electric power training network k_gl Is the k _ g row, k _ l column element, Z of the node impedance matrix k_ll Is the k _ l row and the k _ l column elements of the node impedance matrix;
the electric power training network is arranged at a generator node k _ g and a load node k _ lTransmission capacity between
The calculation process of (2) is shown in equation (6):
in the formula (6), the reaction mixture is,
the maximum transmission capacity of a certain transmission line k _ b of the kth independent power grid in the electric power training network,
is the power transmission distribution factor of the transmission line k _ b relative to the generator node k _ g and the load node k _ l.
Further, the feature data extraction algorithm selects num C X 1 asymmetric convolution kernel, num P X 1 asymmetric pooling Window, convolution step S C The maximum pooling step length is S P After the characteristic information of the electrical characteristic matrix including the power transmission distribution factor matrix, the electrical distance matrix and the maximum transmission capacity matrix is extracted, the dimension of each matrix of the kth independent power network in the corresponding power training network is n k X v, each resulting matrix dimension corresponding to a single independent power network in the power test network is n x v, where the row vector dimension v = n min ×a％，n min Number of nodes N for N independent power networks in a power training network 1 ,...n k ...n N The minimum value of n and the number of nodes of a single independent power network in the power test network, i.e. n min ＝min{{n 1 ,...n k ...n N },n}。
Extracting the feature data to obtain n dimension k And normalizing the matrix of x v and n x v to obtain the electric abstract characteristic matrix of the power network node with uniform row vector dimension and data normalization. Extracting power network node electricityAnd elements in the row vectors of the gas abstract feature matrix are used as the electric feature labels of each node of the independent electric power network and the electric power test network in the corresponding electric power training network.
Further, each item of data in the data set to be normalized includes a network efficiency reduction rate after any node in the independent power network is removed, and the data set to be normalized is normalized by a z-score normalization method, wherein the normalization processing is performed by using a power transmission distribution factor abstract feature matrix, an electrical distance abstract feature matrix, a maximum transmission capacity abstract feature matrix, a node value, a node betweenness centrality, a node near centrality, a node feature vector centrality, and a node clustering coefficient, and is shown in formula (7):
in the formula (7), X i For the ith element of a certain item of data X in the data set to be normalized,
is X i The data after the normalization processing is carried out,
is the average value, N, of a certain item of data X in the data set to be normalized X Is the element number, sigma, of a certain item of data X in the data set to be normalized X The standard deviation of a certain item of data X in the data set to be normalized is obtained;
the method for normalizing the node types in the data set to be normalized is shown as the formula (8):
in the formula (8), V i * And normalizing the node type in the data set to be normalized.
Further, the input parameters of the input layer of the BP neural network model are 6+3v units, wherein the input parameters comprise a node value, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficients, node types and 3v node electrical feature labels; the dimension unification is achieved by taking each node electrical characteristic label of all the power networks as an input parameter of the neural network, and the node electrical characteristic labels of a plurality of power networks can be simultaneously extracted to perform cross-network prediction.
The output layer of the BP neural network model selects a purelin function as a transfer function;
the hidden layer of the BP neural network model selects a tansig function as a transfer function;
considering that the electrical feature labels of 3v nodes extracted from the node electrical abstract feature matrix obtained by convolution pooling are main electrical characteristic input parameters in the neural network, determining the number h of hidden layer neural nodes of the BP neural network model can be calculated by adopting formula (9):
in the formula (9), o is the number of output layer units of the BP neural network model, a is a constant between 1 and 10, and the optimal a is selected to minimize the error after multiple trial and error;
the BP neural network model selects a learngdm function as a learning function, the learngdm function is a gradient descent momentum learning function, and the change rate of the weight or the threshold is calculated by utilizing the input and the error of the neuron, the learning rate of the weight and the threshold and the momentum constant.
Further, inputting the power test network into the constructed BP neural network training model, obtaining the test result of the network efficiency reduction rate after the power test network node is removed, and performing inverse normalization on the test result to obtain the test value of the network efficiency reduction rate after the node subjected to inverse normalization processing is removed
The method of (2) is represented by the formula (10):
in the formula (10), σ DR X in the expression (7) is a standard deviation obtained by the term of the network efficiency reduction rate after the node removal,
for the test results of the network efficiency degradation rate after node removal,
x in expression (7) is an average value obtained by the term of the network efficiency reduction rate after the node removal.
Further, the power test network node vulnerability set is
Compared with the prior art, the invention has the beneficial effects that:
(1) The machine learning algorithm is adopted to carry out data processing analysis on the big data of the power grid, a novel power network vulnerability assessment system is established, and the assessment system is suitable for high-dimensional nonlinear large-scale complex power grid vulnerability assessment and meets the requirement of rapidness and accuracy;
(2) Establishing a characteristic data extraction algorithm by adopting an asymmetric convolution kernel and a pooling window in machine learning, abstracting useful information of a power transmission distribution factor matrix, an electrical distance matrix and a maximum transmission capacity matrix, obtaining an electrical abstract characteristic matrix of nodes of the power network, extracting node electrical characteristic labels of all the power networks, enabling the labels to have uniform dimensionality and serve as neural network input parameters for cross-network prediction, and having universality for all the power networks;
(3) The topological structure and the electrical characteristics of the power network are comprehensively considered by the input parameters and the output parameters of the constructed BP neural network model, the key protection is applied to the obtained priority protection node set, and the safe and stable operation efficiency of the power system can be effectively improved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a block diagram of a BP neural network model constructed in accordance with the present invention;
FIG. 3 is a graph comparing the test values and the actual values of the network efficiency degradation rate after the removal of the nodes of the IEEE118 node test system.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived from the embodiments of the present invention by a person skilled in the art without any creative effort, should be included in the protection scope of the present invention.
The purpose of the invention can be realized by the following technical scheme:
as shown in fig. 1, the method for evaluating the vulnerability of the power network node based on the BP neural network specifically includes the following steps:
s1: respectively building an electric power training network topological structure model comprising a plurality of independent electric power networks and an electric power testing network topological structure model comprising a single independent electric power network;
s2: extracting network topology characteristics of each independent power network in the power training network and network topology characteristics of the power testing network based on a complex network theory, wherein the network topology characteristics comprise node values, node betweenness centrality, node approximate centrality, node feature vector centrality, node clustering coefficients and node types;
s3: acquiring electrical characteristic matrixes in each independent power network and the power test network in the power training network, wherein the electrical characteristic matrixes comprise a power transmission distribution factor matrix, an electrical distance matrix and a maximum transmission capacity matrix;
s4: calculating the network efficiency of each independent power network in the power training network, and calculating the network efficiency reduction rate after any node in the independent power networks is removed;
s5: adopting a convolution kernel and a pooling window in an asymmetric form in machine learning, and constructing a feature data extraction algorithm according to the dimension of the electrical feature matrix of each independent power network and each power test network in the power training network;
s6: acquiring a row vector dimension unification electrical characteristic matrix of each independent power network and each independent power test network in the power training network according to the characteristic data extraction algorithm;
s7: normalizing the vector dimension unified electrical characteristic matrix to obtain an electrical network node electrical abstract characteristic matrix corresponding to each independent power network and the power test network with unified row vector dimension and data normalization, wherein the electrical abstract characteristic matrix of the power network node comprises a power transmission distribution factor abstract characteristic matrix, an electrical distance abstract characteristic matrix and a maximum transmission capacity abstract characteristic matrix;
s8: extracting elements in row vectors of the electrical abstract feature matrix of the power network nodes, and defining the elements as node electrical feature labels in each independent power network and power test network in the power training network;
s9: constructing node characteristic attributes of each independent power network and each power test network in the power training network according to network topology characteristics and node electrical characteristic labels of each independent power network and each power test network in the power training network;
s10: removing any node in the independent power network in the S4 to obtain the network efficiency reduction rate, wherein the electrical abstract feature matrix of the power network node in the S7 comprises a power transmission distribution factor abstract feature matrix, an electrical distance abstract feature matrix and a maximum transmission capacity abstract feature matrix, and the network topology characteristics in the node feature attribute in the S9 comprise a node value, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficients and a node type to form a data set to be normalized;
s11: normalizing each item of data in the data set to be normalized in S10;
s12: as shown in fig. 2, a BP neural network training model is built, the BP neural network includes an input layer, a hidden layer and an output layer, and the specific steps of building the BP neural network training model include:
s121: determining the number of neurons of each layer of an input layer, a hidden layer and an output layer in the BP neural network;
s122: taking the node electrical feature labels and the normalized node feature attributes in the S11, including node values, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficients and node types as input parameters of a BP neural network input layer;
s123: taking the normalized network efficiency reduction rate of the independent power network where the node is located after the node is removed in the S11 as an output parameter of the BP neural network output layer;
s124: inputting the training set into a BP neural network, establishing nonlinear mapping between the characteristic attribute of the input layer power network node and the network efficiency reduction rate after the output layer node is removed, and completing the establishment of a BP neural network training model;
s13: inputting the power test network into the BP neural network training model established in S12, obtaining a test result of the network efficiency reduction rate after the power test network node is removed, and performing reverse normalization on the test result to obtain a test value of the network efficiency reduction rate after the node is removed after the reverse normalization processing;
s14: and establishing a power test network node vulnerability set according to the test value of the network efficiency reduction rate of the power test network, performing descending sorting according to the network efficiency reduction rate of all elements in the node vulnerability set, adding the nodes with the top rank to a key protection node set to obtain a priority protection node set, and applying key protection to the nodes in the set.
Abstracting generators, loads and substations in the electric power training network topological structure model and the electric power test network topological structure model as nodes in the network topological structure model, abstracting power transmission lines and transformer branches in the electric power training network topological structure model and the electric power test network topological structure model as edges in the network topological structure model, wherein the electric power training network topological structure model can be expressed as:
in the formula (1), the reaction mixture is,
n independent power networks are selected as sample data in the power training network, the power training network topological structure model comprises N independent power network topological structure models,
representing a set of all node types in the kth independent grid in the power training network topology model, including a generator node set
Load node set
Transformer substation node set
Representing all edge sets in the power training network topological structure model;
the power test network topology model can be represented as:
G P ＝(V P ,E P ) (2)
in the formula (1), G p Representing the selection of a single independent power network as sample data, V, in a power test network P Representing power test network topologyAll node type sets in the structure model, including generator node set
Load node set
Transformer substation node set V t P ，E P Representing all edge sets in the power test network topology model.
Calculating the network efficiency of the kth independent power grid according to the network efficiency of each independent power network in the electric power training network
For example, the calculation process is shown in formula (3):
in the formula (3), the reaction mixture is,
for the total number of kth individual grid generator nodes in the power training network,
for the kth individual grid load node total in the power training network,
for the electrical distance between the generator node k _ g and the load node k _ l in the kth individual grid in the electrical training network,
the transmission capacity between the generator node k _ g and the load node k _ l for the power training network;
calculating a network efficiency reduction rate after removing any node in the independent power network to remove a node i from the kth independent power networkRate of decrease in collateral efficiency
For example, the calculation process is shown in formula (4):
in the formula (4), the reaction mixture is,
to remove the network efficiency of the independent power network in which the node k i is located,
the larger the node k _ i is, the larger the influence on the power network where the node k _ i is located is, and the more important the node is.
The calculation process of the electrical distance between the generator node k _ g and the load node k _ l in the kth independent power grid in the electric power training network is shown as the formula (5):
in the formula (5), Z k_gg Is the k-th independent power grid node impedance matrix k _ g row and k _ g column elements, Z in the electric power training network k_gl Is the k _ g row, k _ l column element, Z of the node impedance matrix k_ll Is the k _ l row and the k _ l column elements of the node impedance matrix;
the transmission capacity of the electric power training network between the generator node k _ g and the load node k _ l
The calculation process of (2) is shown in equation (6):
in the formula (6), the reaction mixture is,
the maximum transmission capacity of a certain transmission line k _ b of the kth independent power grid in the electric power training network,
is the power transmission distribution factor of the transmission line k _ b relative to the generator node k _ g and the load node k _ l.
Feature data extraction algorithm selects num C X 1 asymmetric convolution kernel, num P X 1 asymmetric pooling Window, convolution step S C The maximum pooling step length is S P After the characteristic information of the electrical characteristic matrix including the power transmission distribution factor matrix, the electrical distance matrix and the maximum transmission capacity matrix is extracted, the dimension of each matrix of the kth independent power network in the corresponding power training network is n k X v, each resulting matrix dimension corresponding to a single independent power network in the power test network is n x v, where the row vector dimension v = n min ×a％，n min Number of nodes N for N independent power networks in a power training network 1 ,...n k ...n N N is the minimum value of the number of nodes n of a single independent power network in the power test network min ＝min{{n 1 ,...n k ...n N N }. The electric power training network adopted by the invention is an IEEE300 node test system and an Italian power grid, and the electric power test network is an IEEE118 node test system, so that the minimum node number of the independent electric power network is the IEEE118 node test system, namely n min =118; through multiple tests, the invention finds that when the a% =60% and the v = 70% are preferably selected, the electrical characteristic data can be better extracted by performing convolution pooling on the electrical characteristic matrix, and the input data quality of the BP neural network model is improved.
The dimension after characteristic data extraction is n k And normalizing the matrix of x v and n x v to obtain the electric abstract characteristic matrix of the power network node with uniform row vector dimension and data normalization. Extracting electricityAnd elements in the row vectors of the electrical abstract feature matrix of the force network nodes are used as electrical feature labels of each node of the independent power network and the power test network in the corresponding power training network.
Each item of data in the data set to be normalized comprises a network efficiency reduction rate after any node in the independent power network is removed, and a power transmission distribution factor abstract characteristic matrix, an electrical distance abstract characteristic matrix, a maximum transmission capacity abstract characteristic matrix, a node value, node betweenness centrality, node nearness centrality, node eigenvector centrality and a node clustering coefficient are normalized by adopting a z-score standardization method, wherein the formula (7) is as follows:
in the formula (7), X i For the ith element of a certain item of data X in the data set to be normalized,
is X i The data after the normalization processing is carried out,
is the average value, N, of a certain item of data X in the data set to be normalized X For the number of elements, σ, of a certain item of data X in the data set to be normalized X The standard deviation of a certain item of data X in the data set to be normalized is obtained;
the method for normalizing the node types in the data set to be normalized is shown as the formula (8):
in the formula (8), V i * And normalizing the node type in the data set to be normalized.
The input parameters of the input layer of the BP neural network model are 6+3v units, wherein the input parameters comprise node value, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficient, node type and 3v node electrical feature labels; the dimension unification is achieved by taking each node electrical characteristic label of all the power networks as an input parameter of the neural network, and the node electrical characteristic labels of a plurality of power networks can be simultaneously extracted to perform cross-network prediction.
The output layer of the BP neural network model selects a purelin function as a transfer function;
the hidden layer of the BP neural network model selects a tansig function as a transfer function;
considering that the electrical feature labels of 3v nodes extracted from the node electrical abstract feature matrix obtained by convolution pooling are main electrical characteristic input parameters in the neural network, determining the number h of hidden layer neural nodes of the BP neural network model can be calculated by adopting formula (9):
in the formula (9), o is the number of output layer units of the BP neural network model, o =1,a is a constant between 1 and 10, the optimal a is selected to minimize the error after multiple trial and error, and the optimal a is preferably selected to be a =4;
the BP neural network model selects a learngdm function as a learning function, the learngdm function is a gradient descent momentum learning function, and the change rate of the weight or the threshold is calculated by utilizing the input and the error of the neuron, the learning rate of the weight and the threshold and the momentum constant.
Inputting the power test network into the constructed BP neural network training model, obtaining the test result of the network efficiency reduction rate after the power test network node is removed, carrying out reverse normalization on the test result, and obtaining the test value of the network efficiency reduction rate after the node subjected to reverse normalization processing is removed
The process of (1) is as in formula (10)The following steps:
in the formula (10), σ DR In the expression (7), X is a standard deviation obtained by taking the term of the network efficiency reduction rate after the node removal,
for the test results of the network efficiency degradation rate after node removal,
in expression (7), X is an average value obtained by taking the term of the network efficiency reduction rate after the node removal.
The invention adopts an electric power training network as an IEEE300 node test system and an Italian power grid, 821 nodes in two independent electric power networks are used as a training set of a BP neural network model, and the electric power testing network is an IEEE118 node test system, namely a test set of the BP neural network model;
the ratio of the training set to the test set of the neural network model is 87.5% to 12.5%.
Fig. 3 shows the test results of the BP neural network on the power test network, i.e. the test values and the actual values of the network efficiency reduction rate after the nodes of the IEEE118 node test system are removed. Five numerical value scoring indexes of root mean square error, average absolute percentage error, mean difference and decision coefficient R square are adopted to measure the testing precision, as shown in table 1, the error between a testing value and a true value is known to be small, and the testing of the network efficiency reduction rate of the built BP neural network model after the nodes of the power testing network are removed is reliable.
TABLE 1
The invention establishes power according to the test value of the network efficiency reduction rate after the node of the IEEE118 node test system is removedForce network node vulnerability aggregation
And sorting in a descending order, and adding the nodes with the top 20 ranks to the key protection node set.
As shown in table 2, the results are sorted for vulnerability of the set of heavy protection nodes in the IEEE118 node test system. Compared with the node at the top 20 of the node importance ranking table in the document [1], the key protection node set obtained by the invention has 12 same nodes; compared with the node at the top 30 of the key node sorting table in the document [2], the key protection node set obtained by the invention has 15 same nodes, and the importance of the node 80, the node 49 and the node 77 is in the position interval with the top sorting in the document [2] and the invention. The result shows that the method provided by the invention can accurately evaluate the node vulnerability and can be effectively applied to the safety protection of the power system.
TABLE 3
Reference documents:
[1] wang Yi, zou Yanli, yellow plum, et al. Grid key node identification that considers both local and global characteristics in combination [ J ]. Compute physics, 2018,35 (1): 8.
[2]Beyza J,Garcia-Paricio E,Ruiz H F,et al.Geodesic Vulnerability Approach for Identification of Critical Buses in Power Systems[J].Journal of Modern Power Systems and Clean Energy,2021,9(01):37-45。
Claims (9)
1. The method for evaluating the vulnerability of the power network node based on the BP neural network is characterized by specifically comprising the following steps of:
s1: respectively building an electric power training network topological structure model comprising a plurality of independent electric power networks and an electric power testing network topological structure model comprising a single independent electric power network;
s2: extracting network topology characteristics of each independent power network in the power training network and network topology characteristics of the power testing network based on a complex network theory, wherein the network topology characteristics comprise node values, node betweenness centrality, node approximate centrality, node feature vector centrality, node clustering coefficients and node types;
s3: acquiring electrical characteristic matrixes in each independent power network and the power test network in the power training network, wherein the electrical characteristic matrixes comprise a power transmission distribution factor matrix, an electrical distance matrix and a maximum transmission capacity matrix;
s4: calculating the network efficiency of each independent power network in the power training network, and calculating the network efficiency reduction rate after any node in the independent power networks is removed;
s5: adopting a convolution kernel and a pooling window in an asymmetric form in machine learning, and constructing a feature data extraction algorithm according to the dimension of the electrical feature matrix of each independent power network and each power test network in the power training network;
s6: acquiring a row vector dimension unified electrical feature matrix of each independent power network and each power test network in the power training network according to the feature data extraction algorithm;
s7: normalizing the vector dimension unified electrical characteristic matrix to obtain an electrical network node electrical abstract characteristic matrix corresponding to each independent power network and the power test network with unified row vector dimension and data normalization, wherein the electrical abstract characteristic matrix of the power network node comprises a power transmission distribution factor abstract characteristic matrix, an electrical distance abstract characteristic matrix and a maximum transmission capacity abstract characteristic matrix;
s8: extracting elements in row vectors of the electrical abstract feature matrix of the power network nodes, and defining the elements as node electrical feature labels in each independent power network and power test network in the power training network;
s9: according to network topological characteristics and node electrical characteristic labels of each independent power network and each power test network in the power training network, node characteristic attributes of each independent power network and each power test network in the power training network are constructed;
s10: removing any node in the independent power network in the S4 to obtain the network efficiency reduction rate, wherein the electrical abstract feature matrix of the power network node in the S7 comprises a power transmission distribution factor abstract feature matrix, an electrical distance abstract feature matrix and a maximum transmission capacity abstract feature matrix, and the network topology characteristics in the node feature attribute in the S9 comprise a node value, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficients and a node type to form a data set to be normalized;
s11: normalizing each item of data in the data set to be normalized in S10;
s12: the method comprises the following steps of building a BP neural network training model, wherein the BP neural network comprises an input layer, a hidden layer and an output layer, and the concrete steps of building the BP neural network training model comprise:
s121: determining the number of neurons of each layer of an input layer, a hidden layer and an output layer in the BP neural network;
s122: taking the node electrical feature labels and the normalized node feature attributes in the S11, including node values, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficients and node types as input parameters of a BP neural network input layer;
s123: taking the normalized network efficiency reduction rate of the independent power network where the node is located after the node is removed in the S11 as an output parameter of the BP neural network output layer;
s124: inputting the training set into a BP neural network, establishing nonlinear mapping between the characteristic attribute of the input layer power network node and the network efficiency reduction rate after the output layer node is removed, and completing the establishment of a BP neural network training model;
s13: inputting the power test network into the BP neural network training model established in S12, obtaining a test result of the network efficiency reduction rate after the power test network node is removed, and performing reverse normalization on the test result to obtain a test value of the network efficiency reduction rate after the node is removed after the reverse normalization processing;
s14: and establishing a power test network node vulnerability set according to the test value of the network efficiency reduction rate of the power test network, performing descending sorting according to the network efficiency reduction rate of all elements in the node vulnerability set, adding the nodes with the top rank to a key protection node set to obtain a priority protection node set, and applying key protection to the nodes in the set.
2. The method for assessing the vulnerability of the power network nodes based on the BP neural network as claimed in claim 1, wherein the generator, the load and the transformer substation in the power training network topology model and the power testing network topology model are abstracted as the nodes in the network topology model, the transmission line and the transformer branch in the power training network topology model and the power testing network topology model are abstracted as the edges in the network topology model, and the power training network topology model can be expressed as:
in the formula (1), the reaction mixture is,
n independent power networks are selected as sample data in the power training network, the power training network topological structure model comprises N independent power network topological structure models,
set representing all node types in the kth independent power grid in the power training network topology model, including generator node set
Load node set
Transformer substation node set
Representing all edge sets in the topological structure model of the electric power training network;
the power test network topology model can be represented as:
G P ＝(V P ,E P ) (2)
in the formula (1), G p Representing the selection of a single independent power network as sample data, V, in a power test network P Representing a set of all node types in a power test network topology model, including a set of generator nodes
Load node set V l P Node set V of transformer substation t P ，E P Representing all edge sets in the power test network topology model.
3. The BP neural network-based power network node vulnerability assessment method according to claim 2, wherein the network efficiency of each independent power network in the power training network is calculated as the network efficiency of the kth independent power grid
For example, the calculation process is shown in formula (3):
in the formula (3), the reaction mixture is,
for the total number of kth individual grid generator nodes in the power training network,
for the kth individual grid load node total in the power training network,
for the electrical distance between the generator node k _ g and the load node k _ l in the kth individual grid in the electrical training network,
the transmission capacity between the generator node k _ g and the load node k _ l for the power training network;
calculating a network efficiency reduction rate after removing any node in the independent power network to obtain a network efficiency reduction rate after removing the node i from the kth independent power network
For example, the calculation process is shown in formula (4):
in the formula (4), the reaction mixture is,
to remove the network efficiency of the independent power network where the node k _ i is located,
the larger the node k _ i is, the more influence on the power network where the node k _ i is located is, and the more important the node is.
4. The BP neural network-based power network node vulnerability assessment method according to claim 3, wherein the calculation process of the electrical distance between the generator node k _ g and the load node k _ l in the kth independent grid in the power training network is as shown in formula (5):
in the formula (5), Z k_gg Is the k-th independent grid node impedance matrix k _ g row and k _ g column elements, Z in the electric power training network k_gl Is the k _ g row, k _ l column element, Z of the node impedance matrix k_ll Is the k _ l row and the k _ l column elements of the node impedance matrix;
the transmission capacity of the electric power training network between the generator node k _ g and the load node k _ l
The calculation process of (2) is shown in equation (6):
in the formula (6), the reaction mixture is,
the maximum transmission capacity of a certain transmission line k _ b of the kth independent power grid in the electric power training network,
is the power transmission distribution factor of the transmission line k _ b relative to the generator node k _ g and the load node k _ l.
5. The BP neural network-based power network node vulnerability assessment method according to claim 4, wherein the feature data extraction algorithm selects num C X 1 asymmetric convolution kernel, num P Asymmetric pooling window of x 1, convolution step S C The maximum pooling step length is S P To what is calledThe electrical characteristic matrix comprises a power transmission distribution factor matrix, an electrical distance matrix and a maximum transmission capacity matrix, and after characteristic information is extracted, the obtained dimension of each matrix of the kth independent power network in the corresponding power training network is n k X v, each resulting matrix dimension corresponding to a single independent power network in the power test network is n x v, where the row vector dimension v = n min ×a％，n min Number of nodes N for N independent power networks in a power training network 1 ,...n k ...n N N is the minimum value of the number of nodes n of a single independent power network in the power test network min ＝min{{n 1 ,...n k ...n N },n}。
6. The method for evaluating the vulnerability of the power network nodes based on the BP neural network as claimed in claim 5, wherein each item of data in the data set to be normalized includes a network efficiency reduction rate after removing any node in the independent power network, a power transmission distribution factor abstract feature matrix, an electrical distance abstract feature matrix, a maximum transmission capacity abstract feature matrix, a node value, a node betweenness centrality, a node approaching centrality, a node feature vector centrality, and a node clustering coefficient, and is normalized by a z-score normalization method, as shown in formula (7):
in the formula (7), X i For the ith element of a certain item of data X in the data set to be normalized,
is X i The data after the normalization processing is carried out,
is the average value, N, of a certain item of data X in the data set to be normalized X For the number of elements, σ, of a certain item of data X in the data set to be normalized X The standard deviation of a certain item of data X in the data set to be normalized is obtained;
the method for normalizing the node types in the data set to be normalized is shown as the formula (8):
in the formula (8), V i * And normalizing the node type in the data set to be normalized.
7. The method for evaluating the vulnerability of the power network nodes based on the BP neural network as claimed in claim 6, wherein the input parameters of the input layer of the BP neural network model have 6+3v units, which include node value, node betweenness centrality, node approaching centrality, node feature vector centrality, node clustering coefficient, node type and 3v node electrical feature labels;
the output layer of the BP neural network model selects a purelin function as a transfer function;
the hidden layer of the BP neural network model selects a tansig function as a transfer function;
determining the number h of hidden layer neural nodes of the BP neural network model may be calculated using equation (9):
in the formula (9), o is the number of output layer units of the BP neural network model, a is a constant between 1 and 10, and the optimal a is selected to minimize the error after multiple trial and error;
and the BP neural network model selects a Learndm function as a learning function.
8. The BP neural network-based power of claim 7The network node vulnerability assessment method is characterized in that the test result is subjected to inverse normalization to obtain a test value of the network efficiency reduction rate after the nodes subjected to inverse normalization processing are removed
The method of (2) is represented by the formula (10):
in the formula (10), σ DR X in the expression (7) is a standard deviation obtained by the term of the network efficiency reduction rate after the node removal,
for the test results of the network efficiency degradation rate after node removal,
in expression (7), X is an average value obtained by taking the term of the network efficiency reduction rate after the node removal.
9. The BP neural network-based power network node vulnerability assessment method according to claim 8, wherein the power test network node vulnerability sets are
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211396612.8A CN115659833A (en) | 2022-11-09 | 2022-11-09 | Power network node vulnerability assessment method based on BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211396612.8A CN115659833A (en) | 2022-11-09 | 2022-11-09 | Power network node vulnerability assessment method based on BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115659833A true CN115659833A (en) | 2023-01-31 |
Family
ID=85016939
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211396612.8A Pending CN115659833A (en) | 2022-11-09 | 2022-11-09 | Power network node vulnerability assessment method based on BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115659833A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116757534A (en) * | 2023-06-15 | 2023-09-15 | 中国标准化研究院 | Intelligent refrigerator reliability analysis method based on neural training network |
| CN117575308A (en) * | 2023-10-17 | 2024-02-20 | 中科宏一教育科技集团有限公司 | Risk assessment method, device and equipment for distributed power distribution network and storage medium |
| CN117575308B (en) * | 2023-10-17 | 2024-06-28 | 中科宏一教育科技集团有限公司 | Risk assessment method, device and equipment for distributed power distribution network and storage medium |
-
2022
- 2022-11-09 CN CN202211396612.8A patent/CN115659833A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116757534A (en) * | 2023-06-15 | 2023-09-15 | 中国标准化研究院 | Intelligent refrigerator reliability analysis method based on neural training network |
| CN116757534B (en) * | 2023-06-15 | 2024-03-15 | 中国标准化研究院 | Intelligent refrigerator reliability analysis method based on neural training network |
| CN117575308A (en) * | 2023-10-17 | 2024-02-20 | 中科宏一教育科技集团有限公司 | Risk assessment method, device and equipment for distributed power distribution network and storage medium |
| CN117575308B (en) * | 2023-10-17 | 2024-06-28 | 中科宏一教育科技集团有限公司 | Risk assessment method, device and equipment for distributed power distribution network and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110175386B (en) | Method for predicting temperature of electrical equipment of transformer substation | |
| CN110417011B (en) | Online dynamic security assessment method based on mutual information and iterative random forest | |
| CN109818349B (en) | Power grid robust state prediction method based on multidimensional state matrix sliding matching | |
| CN110034581B (en) | Interval electrical medium number vulnerability assessment method for power system under wind power grid-connected condition | |
| CN112149873B (en) | Low-voltage station line loss reasonable interval prediction method based on deep learning | |
| CN106384302B (en) | Power distribution network reliability evaluation method and system | |
| CN113536509B (en) | Micro-grid topology identification method based on graph convolution network | |
| CN107133684A (en) | A kind of random matrix construction method towards GA for reactive power optimization | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN111460374A (en) | Power distribution network D-PMU optimal configuration method considering node differences | |
| CN115659833A (en) | Power network node vulnerability assessment method based on BP neural network | |
| CN109066651B (en) | Method for calculating limit transmission power of wind power-load scene | |
| CN104077493A (en) | Method for constructing state evaluation index system of electric relaying protection system | |
| CN110688809A (en) | Box transformer substation fault diagnosis method based on VPRS-RBF neural network | |
| CN111401792A (en) | Dynamic safety assessment method based on extreme gradient lifting decision tree | |
| CN114116832A (en) | Power distribution network abnormity identification method based on data driving | |
| CN113722868A (en) | Multi-index power grid node vulnerability assessment method fusing structure hole characteristics | |
| CN113486291A (en) | Petroleum drilling machine micro-grid fault prediction method based on deep learning | |
| CN109684749B (en) | Photovoltaic power station equivalent modeling method considering operating characteristics | |
| CN115983714A (en) | Static security assessment method and system for edge graph neural network power system | |
| CN113627655B (en) | Method and device for simulating and predicting pre-disaster fault scene of power distribution network | |
| CN111628531B (en) | Data driving method for static voltage stability evaluation of power system | |
| Zhao et al. | An adaptive online learning model for flight data cluster analysis | |
| Ji et al. | Cost Prediction of Distribution Network Project Based on DART Model | |
| Sun et al. | DTformer: An efficient digital twin model for loss measurement in UHVDC transmission systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |